Throughout the election, the USC Dornsife/LA Times Presidential Election Poll has been an outlier, showing Trump more favorably than other polls, often by substantial margins.  It uses a radically different design and is intended in particular to show trends.  The poll describes itself as:

… [a] pioneering approach to tracking changes in Americans' opinions throughout a campaign for the White House. Around 3000 respondents in our representative panel are asked questions on a regular basis on what they care about most in the election, and on their attitudes toward their preferred candidates. The "Daybreak poll" is updated just after midnight every day of the week.

The poll this morning shows that Trump has reversed his decline and is ahead by a statistically insignificant fraction of one percent, 44.4% to 43.8%.  But look at the trends:

The consensus throughout the media rooting for Hillary to win (95% of them) is that Trump is over, done, and needing a fork stuck in him as soon as possible.  But nobody has a clue as to what turnout will be this time.  Nobody knows how much vote fraud there will be.

If unenthusiastic Hillary supporters believe she will win, they may not bother to vote.  And if new voters flock to support Trump, the turnout models used by he polls, based on past experience, may be worthless.

Nobody knows, and we have 18 more days before Election Day.